Parakeet by Nvidia

Parakeet delivers state-of-the-art speech recognition through Nvidia's open-source models optimized for GPU acceleration. Researchers, developers, and enterprises deploy locally eliminating API costs and latency. Supports English with multilingual expansion planned. Automatic punctuation and capitalization create publication-ready transcripts. Word-level timestamps enable precise audio-text alignment. RNNT architecture handles long audio sequences efficiently. Multiple model sizes balance accuracy and speed. Hugging Face integration simplifies deployment. Docker containers enable production scaling. Fine-tuning scripts support domain adaptation. Commercial license permits enterprise applications. Completely free open-source weights. Runs locally on Nvidia GPUs. Accuracy rivals cloud services without vendor dependency.